Intelligent fiber co-trench identification algorithm and experiment

The coexistence of primary and backup optical fibers in the same trench poses a high-risk event due to the passive nature of optical fibers, which may lead to the sharing of the same trench by primary and backup services. When faults occur, primary and backup services will interrupt at the same time. Currently, it is difficult and inefficient to rely solely on manual identification of co-trench fibers, which brings significant risks to the service quality of communication networks. To enhance the survivability of communication networks, this paper proposes and validates an ensemble learning detection architecture, utilizing artificial intelligence to identify co-trench fibers. This architecture achieves high-precision online identification of co-trench fibers in operator networks without the need for manual tapping for detection along the routes. This paper employs the curve similarity learner based on the dynamic characteristics of fiber vibration for co-trench fiber identification. For the first time, field trials on co-trench fiber identification were conducted. Preprocessing of dynamic fiber vibration data collected from the field is performed, and evaluation is conducted based on metrics such as accuracy and <roman>F</roman>1-score derived from the confusion matrix. The curve similarity ensemble learning algorithm proposed in this paper improves accuracy by 6.1% compared to sub-optimal non-ensemble learning models. Experimental results validate the reliability and advancement of the proposed architecture.